Truace
Subscribe

Functional Near-Infrared Spectroscopy Combined with Machine Learning for Exploratory Classification of Autism Spectrum Disorder in Young Children

Early identification of autism spectrum disorder (ASD) may facilitate timely intervention, while diagnosis remains clinically based. This exploratory study evaluated fNIRS-derived features using multiple machine-learning algorithms to compare observed classification performance between young children aged 2-6 years with ASD and typically developing (TD) children. β-activation values from eight channels were used as features to differentiate the ASD and TD groups. Two paradigm-specific analytic datasets were anal…

By the Truvace desk2 min read1 sourcePeer-reviewedTRV-2026-2016· cite & verify
Functional Near-Infrared Spectroscopy Combined with Machine Learning for Exploratory Classification of Autism Spectrum Disorder in Young Children
"Mental Patients Against Stigma" by Mindy Fisher is licensed under CC BY 2.0. To view a copy of this license, visit https://creativecommons.org/licenses/by/2.0/.

In brief

Researchers tested whether task-based functional near-infrared spectroscopy signals could help machine-learning models separate children with autism spectrum disorder from typically developing peers aged 2-6. They used beta-activation values from eight channels across two task paradigms with about 60 children each, and compared seven common classifiers using leave-one-out cross-validation and permutation testing.

The result matters because autism diagnosis remains clinically based and earlier identification could allow timely intervention, yet this approach did not yet deliver a usable aid. The work shows the data can be collected and analyzed in practice, but whether fNIRS-based models can add diagnostic value remains unresolved pending larger, sex-balanced external validation.

Main points

  1. Paradigm 1 included 60 children (29 with ASD, 31 typically developing) with 24 beta features each; Paradigm 2 included 58 children with 16 features each.
  2. In Paradigm 1, K-nearest neighbors had the highest observed accuracy (0.567) and decision tree the highest observed AUC (0.582).
  3. Paradigm 2 showed numerically higher performance on several metrics, but no classifier significantly exceeded chance.
  4. The fNIRS features added no statistically significant predictive value beyond age and sex.

The problem

Seven machine-learning classifiers applied to fNIRS features from children aged 2-6 did not significantly exceed chance in permutation tests, so no diagnostic value beyond age and sex was established.

The rundown

The study compared K-nearest neighbors, logistic regression, support vector machine, decision tree, random forest, gradient boosting decision trees, and XGBoost. Performance was judged by accuracy, specificity, sensitivity, and AUC, with normalization and principal component analysis used for feature extraction.

Permutation testing checked whether observed performance beat what chance would produce. Even the best Paradigm 1 figures were modest, and the authors describe the work as exploratory, supporting only the procedural feasibility of acquiring and analyzing task-based fNIRS data.

Sources

  1. Peer-reviewedJournal of Visualized Experiments2026-10-13

The debate